2 * libata-eh.c - libata error handling
4 * Maintained by: Jeff Garzik <jgarzik@pobox.com>
5 * Please ALWAYS copy linux-ide@vger.kernel.org
8 * Copyright 2006 Tejun Heo <htejun@gmail.com>
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public License as
13 * published by the Free Software Foundation; either version 2, or
14 * (at your option) any later version.
16 * This program is distributed in the hope that it will be useful,
17 * but WITHOUT ANY WARRANTY; without even the implied warranty of
18 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
19 * General Public License for more details.
21 * You should have received a copy of the GNU General Public License
22 * along with this program; see the file COPYING. If not, write to
23 * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
27 * libata documentation is available via 'make {ps|pdf}docs',
28 * as Documentation/DocBook/libata.*
30 * Hardware documentation available from http://www.t13.org/ and
31 * http://www.sata-io.org/
35 #include <linux/kernel.h>
36 #include <scsi/scsi.h>
37 #include <scsi/scsi_host.h>
38 #include <scsi/scsi_eh.h>
39 #include <scsi/scsi_device.h>
40 #include <scsi/scsi_cmnd.h>
41 #include "../scsi/scsi_transport_api.h"
43 #include <linux/libata.h>
48 ATA_EH_SPDN_NCQ_OFF = (1 << 0),
49 ATA_EH_SPDN_SPEED_DOWN = (1 << 1),
50 ATA_EH_SPDN_FALLBACK_TO_PIO = (1 << 2),
53 /* Waiting in ->prereset can never be reliable. It's sometimes nice
54 * to wait there but it can't be depended upon; otherwise, we wouldn't
55 * be resetting. Just give it enough time for most drives to spin up.
58 ATA_EH_PRERESET_TIMEOUT = 10 * HZ,
61 /* The following table determines how we sequence resets. Each entry
62 * represents timeout for that try. The first try can be soft or
63 * hardreset. All others are hardreset if available. In most cases
64 * the first reset w/ 10sec timeout should succeed. Following entries
65 * are mostly for error handling, hotplug and retarded devices.
67 static const unsigned long ata_eh_reset_timeouts[] = {
68 10 * HZ, /* most drives spin up by 10sec */
69 10 * HZ, /* > 99% working drives spin up before 20sec */
70 35 * HZ, /* give > 30 secs of idleness for retarded devices */
71 5 * HZ, /* and sweet one last chance */
72 /* > 1 min has elapsed, give up */
75 static void __ata_port_freeze(struct ata_port *ap);
76 static void ata_eh_finish(struct ata_port *ap);
78 static void ata_eh_handle_port_suspend(struct ata_port *ap);
79 static void ata_eh_handle_port_resume(struct ata_port *ap);
81 static void ata_eh_handle_port_suspend(struct ata_port *ap)
84 static void ata_eh_handle_port_resume(struct ata_port *ap)
86 #endif /* CONFIG_PM */
88 static void ata_ering_record(struct ata_ering *ering, int is_io,
89 unsigned int err_mask)
91 struct ata_ering_entry *ent;
96 ering->cursor %= ATA_ERING_SIZE;
98 ent = &ering->ring[ering->cursor];
100 ent->err_mask = err_mask;
101 ent->timestamp = get_jiffies_64();
104 static void ata_ering_clear(struct ata_ering *ering)
106 memset(ering, 0, sizeof(*ering));
109 static int ata_ering_map(struct ata_ering *ering,
110 int (*map_fn)(struct ata_ering_entry *, void *),
114 struct ata_ering_entry *ent;
118 ent = &ering->ring[idx];
121 rc = map_fn(ent, arg);
124 idx = (idx - 1 + ATA_ERING_SIZE) % ATA_ERING_SIZE;
125 } while (idx != ering->cursor);
130 static unsigned int ata_eh_dev_action(struct ata_device *dev)
132 struct ata_eh_context *ehc = &dev->ap->eh_context;
134 return ehc->i.action | ehc->i.dev_action[dev->devno];
137 static void ata_eh_clear_action(struct ata_device *dev,
138 struct ata_eh_info *ehi, unsigned int action)
143 ehi->action &= ~action;
144 for (i = 0; i < ATA_MAX_DEVICES; i++)
145 ehi->dev_action[i] &= ~action;
147 /* doesn't make sense for port-wide EH actions */
148 WARN_ON(!(action & ATA_EH_PERDEV_MASK));
150 /* break ehi->action into ehi->dev_action */
151 if (ehi->action & action) {
152 for (i = 0; i < ATA_MAX_DEVICES; i++)
153 ehi->dev_action[i] |= ehi->action & action;
154 ehi->action &= ~action;
157 /* turn off the specified per-dev action */
158 ehi->dev_action[dev->devno] &= ~action;
163 * ata_scsi_timed_out - SCSI layer time out callback
164 * @cmd: timed out SCSI command
166 * Handles SCSI layer timeout. We race with normal completion of
167 * the qc for @cmd. If the qc is already gone, we lose and let
168 * the scsi command finish (EH_HANDLED). Otherwise, the qc has
169 * timed out and EH should be invoked. Prevent ata_qc_complete()
170 * from finishing it by setting EH_SCHEDULED and return
173 * TODO: kill this function once old EH is gone.
176 * Called from timer context
179 * EH_HANDLED or EH_NOT_HANDLED
181 enum scsi_eh_timer_return ata_scsi_timed_out(struct scsi_cmnd *cmd)
183 struct Scsi_Host *host = cmd->device->host;
184 struct ata_port *ap = ata_shost_to_port(host);
186 struct ata_queued_cmd *qc;
187 enum scsi_eh_timer_return ret;
191 if (ap->ops->error_handler) {
192 ret = EH_NOT_HANDLED;
197 spin_lock_irqsave(ap->lock, flags);
198 qc = ata_qc_from_tag(ap, ap->active_tag);
200 WARN_ON(qc->scsicmd != cmd);
201 qc->flags |= ATA_QCFLAG_EH_SCHEDULED;
202 qc->err_mask |= AC_ERR_TIMEOUT;
203 ret = EH_NOT_HANDLED;
205 spin_unlock_irqrestore(ap->lock, flags);
208 DPRINTK("EXIT, ret=%d\n", ret);
213 * ata_scsi_error - SCSI layer error handler callback
214 * @host: SCSI host on which error occurred
216 * Handles SCSI-layer-thrown error events.
219 * Inherited from SCSI layer (none, can sleep)
224 void ata_scsi_error(struct Scsi_Host *host)
226 struct ata_port *ap = ata_shost_to_port(host);
227 int i, repeat_cnt = ATA_EH_MAX_REPEAT;
232 /* synchronize with port task */
233 ata_port_flush_task(ap);
235 /* synchronize with host lock and sort out timeouts */
237 /* For new EH, all qcs are finished in one of three ways -
238 * normal completion, error completion, and SCSI timeout.
239 * Both cmpletions can race against SCSI timeout. When normal
240 * completion wins, the qc never reaches EH. When error
241 * completion wins, the qc has ATA_QCFLAG_FAILED set.
243 * When SCSI timeout wins, things are a bit more complex.
244 * Normal or error completion can occur after the timeout but
245 * before this point. In such cases, both types of
246 * completions are honored. A scmd is determined to have
247 * timed out iff its associated qc is active and not failed.
249 if (ap->ops->error_handler) {
250 struct scsi_cmnd *scmd, *tmp;
253 spin_lock_irqsave(ap->lock, flags);
255 list_for_each_entry_safe(scmd, tmp, &host->eh_cmd_q, eh_entry) {
256 struct ata_queued_cmd *qc;
258 for (i = 0; i < ATA_MAX_QUEUE; i++) {
259 qc = __ata_qc_from_tag(ap, i);
260 if (qc->flags & ATA_QCFLAG_ACTIVE &&
265 if (i < ATA_MAX_QUEUE) {
266 /* the scmd has an associated qc */
267 if (!(qc->flags & ATA_QCFLAG_FAILED)) {
268 /* which hasn't failed yet, timeout */
269 qc->err_mask |= AC_ERR_TIMEOUT;
270 qc->flags |= ATA_QCFLAG_FAILED;
274 /* Normal completion occurred after
275 * SCSI timeout but before this point.
276 * Successfully complete it.
278 scmd->retries = scmd->allowed;
279 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
283 /* If we have timed out qcs. They belong to EH from
284 * this point but the state of the controller is
285 * unknown. Freeze the port to make sure the IRQ
286 * handler doesn't diddle with those qcs. This must
287 * be done atomically w.r.t. setting QCFLAG_FAILED.
290 __ata_port_freeze(ap);
292 spin_unlock_irqrestore(ap->lock, flags);
294 spin_unlock_wait(ap->lock);
297 /* invoke error handler */
298 if (ap->ops->error_handler) {
299 /* process port resume request */
300 ata_eh_handle_port_resume(ap);
302 /* fetch & clear EH info */
303 spin_lock_irqsave(ap->lock, flags);
305 memset(&ap->eh_context, 0, sizeof(ap->eh_context));
306 ap->eh_context.i = ap->eh_info;
307 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
309 ap->pflags |= ATA_PFLAG_EH_IN_PROGRESS;
310 ap->pflags &= ~ATA_PFLAG_EH_PENDING;
312 spin_unlock_irqrestore(ap->lock, flags);
314 /* invoke EH, skip if unloading or suspended */
315 if (!(ap->pflags & (ATA_PFLAG_UNLOADING | ATA_PFLAG_SUSPENDED)))
316 ap->ops->error_handler(ap);
320 /* process port suspend request */
321 ata_eh_handle_port_suspend(ap);
323 /* Exception might have happend after ->error_handler
324 * recovered the port but before this point. Repeat
327 spin_lock_irqsave(ap->lock, flags);
329 if (ap->pflags & ATA_PFLAG_EH_PENDING) {
331 ata_port_printk(ap, KERN_INFO,
332 "EH pending after completion, "
333 "repeating EH (cnt=%d)\n", repeat_cnt);
334 spin_unlock_irqrestore(ap->lock, flags);
337 ata_port_printk(ap, KERN_ERR, "EH pending after %d "
338 "tries, giving up\n", ATA_EH_MAX_REPEAT);
341 /* this run is complete, make sure EH info is clear */
342 memset(&ap->eh_info, 0, sizeof(ap->eh_info));
344 /* Clear host_eh_scheduled while holding ap->lock such
345 * that if exception occurs after this point but
346 * before EH completion, SCSI midlayer will
349 host->host_eh_scheduled = 0;
351 spin_unlock_irqrestore(ap->lock, flags);
353 WARN_ON(ata_qc_from_tag(ap, ap->active_tag) == NULL);
354 ap->ops->eng_timeout(ap);
357 /* finish or retry handled scmd's and clean up */
358 WARN_ON(host->host_failed || !list_empty(&host->eh_cmd_q));
360 scsi_eh_flush_done_q(&ap->eh_done_q);
363 spin_lock_irqsave(ap->lock, flags);
365 if (ap->pflags & ATA_PFLAG_LOADING)
366 ap->pflags &= ~ATA_PFLAG_LOADING;
367 else if (ap->pflags & ATA_PFLAG_SCSI_HOTPLUG)
368 queue_delayed_work(ata_aux_wq, &ap->hotplug_task, 0);
370 if (ap->pflags & ATA_PFLAG_RECOVERED)
371 ata_port_printk(ap, KERN_INFO, "EH complete\n");
373 ap->pflags &= ~(ATA_PFLAG_SCSI_HOTPLUG | ATA_PFLAG_RECOVERED);
375 /* tell wait_eh that we're done */
376 ap->pflags &= ~ATA_PFLAG_EH_IN_PROGRESS;
377 wake_up_all(&ap->eh_wait_q);
379 spin_unlock_irqrestore(ap->lock, flags);
385 * ata_port_wait_eh - Wait for the currently pending EH to complete
386 * @ap: Port to wait EH for
388 * Wait until the currently pending EH is complete.
391 * Kernel thread context (may sleep).
393 void ata_port_wait_eh(struct ata_port *ap)
399 spin_lock_irqsave(ap->lock, flags);
401 while (ap->pflags & (ATA_PFLAG_EH_PENDING | ATA_PFLAG_EH_IN_PROGRESS)) {
402 prepare_to_wait(&ap->eh_wait_q, &wait, TASK_UNINTERRUPTIBLE);
403 spin_unlock_irqrestore(ap->lock, flags);
405 spin_lock_irqsave(ap->lock, flags);
407 finish_wait(&ap->eh_wait_q, &wait);
409 spin_unlock_irqrestore(ap->lock, flags);
411 /* make sure SCSI EH is complete */
412 if (scsi_host_in_recovery(ap->scsi_host)) {
419 * ata_qc_timeout - Handle timeout of queued command
420 * @qc: Command that timed out
422 * Some part of the kernel (currently, only the SCSI layer)
423 * has noticed that the active command on port @ap has not
424 * completed after a specified length of time. Handle this
425 * condition by disabling DMA (if necessary) and completing
426 * transactions, with error if necessary.
428 * This also handles the case of the "lost interrupt", where
429 * for some reason (possibly hardware bug, possibly driver bug)
430 * an interrupt was not delivered to the driver, even though the
431 * transaction completed successfully.
433 * TODO: kill this function once old EH is gone.
436 * Inherited from SCSI layer (none, can sleep)
438 static void ata_qc_timeout(struct ata_queued_cmd *qc)
440 struct ata_port *ap = qc->ap;
441 u8 host_stat = 0, drv_stat;
446 ap->hsm_task_state = HSM_ST_IDLE;
448 spin_lock_irqsave(ap->lock, flags);
450 switch (qc->tf.protocol) {
453 case ATA_PROT_ATAPI_DMA:
454 host_stat = ap->ops->bmdma_status(ap);
456 /* before we do anything else, clear DMA-Start bit */
457 ap->ops->bmdma_stop(qc);
463 drv_stat = ata_chk_status(ap);
465 /* ack bmdma irq events */
466 ap->ops->irq_clear(ap);
468 ata_dev_printk(qc->dev, KERN_ERR, "command 0x%x timeout, "
469 "stat 0x%x host_stat 0x%x\n",
470 qc->tf.command, drv_stat, host_stat);
472 /* complete taskfile transaction */
473 qc->err_mask |= AC_ERR_TIMEOUT;
477 spin_unlock_irqrestore(ap->lock, flags);
479 ata_eh_qc_complete(qc);
485 * ata_eng_timeout - Handle timeout of queued command
486 * @ap: Port on which timed-out command is active
488 * Some part of the kernel (currently, only the SCSI layer)
489 * has noticed that the active command on port @ap has not
490 * completed after a specified length of time. Handle this
491 * condition by disabling DMA (if necessary) and completing
492 * transactions, with error if necessary.
494 * This also handles the case of the "lost interrupt", where
495 * for some reason (possibly hardware bug, possibly driver bug)
496 * an interrupt was not delivered to the driver, even though the
497 * transaction completed successfully.
499 * TODO: kill this function once old EH is gone.
502 * Inherited from SCSI layer (none, can sleep)
504 void ata_eng_timeout(struct ata_port *ap)
508 ata_qc_timeout(ata_qc_from_tag(ap, ap->active_tag));
514 * ata_qc_schedule_eh - schedule qc for error handling
515 * @qc: command to schedule error handling for
517 * Schedule error handling for @qc. EH will kick in as soon as
518 * other commands are drained.
521 * spin_lock_irqsave(host lock)
523 void ata_qc_schedule_eh(struct ata_queued_cmd *qc)
525 struct ata_port *ap = qc->ap;
527 WARN_ON(!ap->ops->error_handler);
529 qc->flags |= ATA_QCFLAG_FAILED;
530 qc->ap->pflags |= ATA_PFLAG_EH_PENDING;
532 /* The following will fail if timeout has already expired.
533 * ata_scsi_error() takes care of such scmds on EH entry.
534 * Note that ATA_QCFLAG_FAILED is unconditionally set after
535 * this function completes.
537 scsi_req_abort_cmd(qc->scsicmd);
541 * ata_port_schedule_eh - schedule error handling without a qc
542 * @ap: ATA port to schedule EH for
544 * Schedule error handling for @ap. EH will kick in as soon as
545 * all commands are drained.
548 * spin_lock_irqsave(host lock)
550 void ata_port_schedule_eh(struct ata_port *ap)
552 WARN_ON(!ap->ops->error_handler);
554 ap->pflags |= ATA_PFLAG_EH_PENDING;
555 scsi_schedule_eh(ap->scsi_host);
557 DPRINTK("port EH scheduled\n");
561 * ata_port_abort - abort all qc's on the port
562 * @ap: ATA port to abort qc's for
564 * Abort all active qc's of @ap and schedule EH.
567 * spin_lock_irqsave(host lock)
570 * Number of aborted qc's.
572 int ata_port_abort(struct ata_port *ap)
574 int tag, nr_aborted = 0;
576 WARN_ON(!ap->ops->error_handler);
578 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
579 struct ata_queued_cmd *qc = ata_qc_from_tag(ap, tag);
582 qc->flags |= ATA_QCFLAG_FAILED;
589 ata_port_schedule_eh(ap);
595 * __ata_port_freeze - freeze port
596 * @ap: ATA port to freeze
598 * This function is called when HSM violation or some other
599 * condition disrupts normal operation of the port. Frozen port
600 * is not allowed to perform any operation until the port is
601 * thawed, which usually follows a successful reset.
603 * ap->ops->freeze() callback can be used for freezing the port
604 * hardware-wise (e.g. mask interrupt and stop DMA engine). If a
605 * port cannot be frozen hardware-wise, the interrupt handler
606 * must ack and clear interrupts unconditionally while the port
610 * spin_lock_irqsave(host lock)
612 static void __ata_port_freeze(struct ata_port *ap)
614 WARN_ON(!ap->ops->error_handler);
619 ap->pflags |= ATA_PFLAG_FROZEN;
621 DPRINTK("ata%u port frozen\n", ap->print_id);
625 * ata_port_freeze - abort & freeze port
626 * @ap: ATA port to freeze
628 * Abort and freeze @ap.
631 * spin_lock_irqsave(host lock)
634 * Number of aborted commands.
636 int ata_port_freeze(struct ata_port *ap)
640 WARN_ON(!ap->ops->error_handler);
642 nr_aborted = ata_port_abort(ap);
643 __ata_port_freeze(ap);
649 * ata_eh_freeze_port - EH helper to freeze port
650 * @ap: ATA port to freeze
657 void ata_eh_freeze_port(struct ata_port *ap)
661 if (!ap->ops->error_handler)
664 spin_lock_irqsave(ap->lock, flags);
665 __ata_port_freeze(ap);
666 spin_unlock_irqrestore(ap->lock, flags);
670 * ata_port_thaw_port - EH helper to thaw port
671 * @ap: ATA port to thaw
673 * Thaw frozen port @ap.
678 void ata_eh_thaw_port(struct ata_port *ap)
682 if (!ap->ops->error_handler)
685 spin_lock_irqsave(ap->lock, flags);
687 ap->pflags &= ~ATA_PFLAG_FROZEN;
692 spin_unlock_irqrestore(ap->lock, flags);
694 DPRINTK("ata%u port thawed\n", ap->print_id);
697 static void ata_eh_scsidone(struct scsi_cmnd *scmd)
702 static void __ata_eh_qc_complete(struct ata_queued_cmd *qc)
704 struct ata_port *ap = qc->ap;
705 struct scsi_cmnd *scmd = qc->scsicmd;
708 spin_lock_irqsave(ap->lock, flags);
709 qc->scsidone = ata_eh_scsidone;
710 __ata_qc_complete(qc);
711 WARN_ON(ata_tag_valid(qc->tag));
712 spin_unlock_irqrestore(ap->lock, flags);
714 scsi_eh_finish_cmd(scmd, &ap->eh_done_q);
718 * ata_eh_qc_complete - Complete an active ATA command from EH
719 * @qc: Command to complete
721 * Indicate to the mid and upper layers that an ATA command has
722 * completed. To be used from EH.
724 void ata_eh_qc_complete(struct ata_queued_cmd *qc)
726 struct scsi_cmnd *scmd = qc->scsicmd;
727 scmd->retries = scmd->allowed;
728 __ata_eh_qc_complete(qc);
732 * ata_eh_qc_retry - Tell midlayer to retry an ATA command after EH
733 * @qc: Command to retry
735 * Indicate to the mid and upper layers that an ATA command
736 * should be retried. To be used from EH.
738 * SCSI midlayer limits the number of retries to scmd->allowed.
739 * scmd->retries is decremented for commands which get retried
740 * due to unrelated failures (qc->err_mask is zero).
742 void ata_eh_qc_retry(struct ata_queued_cmd *qc)
744 struct scsi_cmnd *scmd = qc->scsicmd;
745 if (!qc->err_mask && scmd->retries)
747 __ata_eh_qc_complete(qc);
751 * ata_eh_detach_dev - detach ATA device
752 * @dev: ATA device to detach
759 static void ata_eh_detach_dev(struct ata_device *dev)
761 struct ata_port *ap = dev->ap;
764 ata_dev_disable(dev);
766 spin_lock_irqsave(ap->lock, flags);
768 dev->flags &= ~ATA_DFLAG_DETACH;
770 if (ata_scsi_offline_dev(dev)) {
771 dev->flags |= ATA_DFLAG_DETACHED;
772 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
775 /* clear per-dev EH actions */
776 ata_eh_clear_action(dev, &ap->eh_info, ATA_EH_PERDEV_MASK);
777 ata_eh_clear_action(dev, &ap->eh_context.i, ATA_EH_PERDEV_MASK);
779 spin_unlock_irqrestore(ap->lock, flags);
783 * ata_eh_about_to_do - about to perform eh_action
784 * @ap: target ATA port
785 * @dev: target ATA dev for per-dev action (can be NULL)
786 * @action: action about to be performed
788 * Called just before performing EH actions to clear related bits
789 * in @ap->eh_info such that eh actions are not unnecessarily
795 static void ata_eh_about_to_do(struct ata_port *ap, struct ata_device *dev,
799 struct ata_eh_info *ehi = &ap->eh_info;
800 struct ata_eh_context *ehc = &ap->eh_context;
802 spin_lock_irqsave(ap->lock, flags);
804 /* Reset is represented by combination of actions and EHI
805 * flags. Suck in all related bits before clearing eh_info to
806 * avoid losing requested action.
808 if (action & ATA_EH_RESET_MASK) {
809 ehc->i.action |= ehi->action & ATA_EH_RESET_MASK;
810 ehc->i.flags |= ehi->flags & ATA_EHI_RESET_MODIFIER_MASK;
812 /* make sure all reset actions are cleared & clear EHI flags */
813 action |= ATA_EH_RESET_MASK;
814 ehi->flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
817 ata_eh_clear_action(dev, ehi, action);
819 if (!(ehc->i.flags & ATA_EHI_QUIET))
820 ap->pflags |= ATA_PFLAG_RECOVERED;
822 spin_unlock_irqrestore(ap->lock, flags);
826 * ata_eh_done - EH action complete
827 * @ap: target ATA port
828 * @dev: target ATA dev for per-dev action (can be NULL)
829 * @action: action just completed
831 * Called right after performing EH actions to clear related bits
832 * in @ap->eh_context.
837 static void ata_eh_done(struct ata_port *ap, struct ata_device *dev,
840 /* if reset is complete, clear all reset actions & reset modifier */
841 if (action & ATA_EH_RESET_MASK) {
842 action |= ATA_EH_RESET_MASK;
843 ap->eh_context.i.flags &= ~ATA_EHI_RESET_MODIFIER_MASK;
846 ata_eh_clear_action(dev, &ap->eh_context.i, action);
850 * ata_err_string - convert err_mask to descriptive string
851 * @err_mask: error mask to convert to string
853 * Convert @err_mask to descriptive string. Errors are
854 * prioritized according to severity and only the most severe
861 * Descriptive string for @err_mask
863 static const char * ata_err_string(unsigned int err_mask)
865 if (err_mask & AC_ERR_HOST_BUS)
866 return "host bus error";
867 if (err_mask & AC_ERR_ATA_BUS)
868 return "ATA bus error";
869 if (err_mask & AC_ERR_TIMEOUT)
871 if (err_mask & AC_ERR_HSM)
872 return "HSM violation";
873 if (err_mask & AC_ERR_SYSTEM)
874 return "internal error";
875 if (err_mask & AC_ERR_MEDIA)
876 return "media error";
877 if (err_mask & AC_ERR_INVALID)
878 return "invalid argument";
879 if (err_mask & AC_ERR_DEV)
880 return "device error";
881 return "unknown error";
885 * ata_read_log_page - read a specific log page
886 * @dev: target device
887 * @page: page to read
888 * @buf: buffer to store read page
889 * @sectors: number of sectors to read
891 * Read log page using READ_LOG_EXT command.
894 * Kernel thread context (may sleep).
897 * 0 on success, AC_ERR_* mask otherwise.
899 static unsigned int ata_read_log_page(struct ata_device *dev,
900 u8 page, void *buf, unsigned int sectors)
902 struct ata_taskfile tf;
903 unsigned int err_mask;
905 DPRINTK("read log page - page %d\n", page);
907 ata_tf_init(dev, &tf);
908 tf.command = ATA_CMD_READ_LOG_EXT;
911 tf.hob_nsect = sectors >> 8;
912 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_LBA48 | ATA_TFLAG_DEVICE;
913 tf.protocol = ATA_PROT_PIO;
915 err_mask = ata_exec_internal(dev, &tf, NULL, DMA_FROM_DEVICE,
916 buf, sectors * ATA_SECT_SIZE);
918 DPRINTK("EXIT, err_mask=%x\n", err_mask);
923 * ata_eh_read_log_10h - Read log page 10h for NCQ error details
924 * @dev: Device to read log page 10h from
925 * @tag: Resulting tag of the failed command
926 * @tf: Resulting taskfile registers of the failed command
928 * Read log page 10h to obtain NCQ error details and clear error
932 * Kernel thread context (may sleep).
935 * 0 on success, -errno otherwise.
937 static int ata_eh_read_log_10h(struct ata_device *dev,
938 int *tag, struct ata_taskfile *tf)
940 u8 *buf = dev->ap->sector_buf;
941 unsigned int err_mask;
945 err_mask = ata_read_log_page(dev, ATA_LOG_SATA_NCQ, buf, 1);
950 for (i = 0; i < ATA_SECT_SIZE; i++)
953 ata_dev_printk(dev, KERN_WARNING,
954 "invalid checksum 0x%x on log page 10h\n", csum);
959 *tag = buf[0] & 0x1f;
961 tf->command = buf[2];
962 tf->feature = buf[3];
967 tf->hob_lbal = buf[8];
968 tf->hob_lbam = buf[9];
969 tf->hob_lbah = buf[10];
971 tf->hob_nsect = buf[13];
977 * atapi_eh_request_sense - perform ATAPI REQUEST_SENSE
978 * @dev: device to perform REQUEST_SENSE to
979 * @sense_buf: result sense data buffer (SCSI_SENSE_BUFFERSIZE bytes long)
981 * Perform ATAPI REQUEST_SENSE after the device reported CHECK
982 * SENSE. This function is EH helper.
985 * Kernel thread context (may sleep).
988 * 0 on success, AC_ERR_* mask on failure
990 static unsigned int atapi_eh_request_sense(struct ata_queued_cmd *qc)
992 struct ata_device *dev = qc->dev;
993 unsigned char *sense_buf = qc->scsicmd->sense_buffer;
994 struct ata_port *ap = dev->ap;
995 struct ata_taskfile tf;
996 u8 cdb[ATAPI_CDB_LEN];
998 DPRINTK("ATAPI request sense\n");
1000 /* FIXME: is this needed? */
1001 memset(sense_buf, 0, SCSI_SENSE_BUFFERSIZE);
1003 /* initialize sense_buf with the error register,
1004 * for the case where they are -not- overwritten
1006 sense_buf[0] = 0x70;
1007 sense_buf[2] = qc->result_tf.feature >> 4;
1009 /* some devices time out if garbage left in tf */
1010 ata_tf_init(dev, &tf);
1012 memset(cdb, 0, ATAPI_CDB_LEN);
1013 cdb[0] = REQUEST_SENSE;
1014 cdb[4] = SCSI_SENSE_BUFFERSIZE;
1016 tf.flags |= ATA_TFLAG_ISADDR | ATA_TFLAG_DEVICE;
1017 tf.command = ATA_CMD_PACKET;
1019 /* is it pointless to prefer PIO for "safety reasons"? */
1020 if (ap->flags & ATA_FLAG_PIO_DMA) {
1021 tf.protocol = ATA_PROT_ATAPI_DMA;
1022 tf.feature |= ATAPI_PKT_DMA;
1024 tf.protocol = ATA_PROT_ATAPI;
1025 tf.lbam = (8 * 1024) & 0xff;
1026 tf.lbah = (8 * 1024) >> 8;
1029 return ata_exec_internal(dev, &tf, cdb, DMA_FROM_DEVICE,
1030 sense_buf, SCSI_SENSE_BUFFERSIZE);
1034 * ata_eh_analyze_serror - analyze SError for a failed port
1035 * @ap: ATA port to analyze SError for
1037 * Analyze SError if available and further determine cause of
1043 static void ata_eh_analyze_serror(struct ata_port *ap)
1045 struct ata_eh_context *ehc = &ap->eh_context;
1046 u32 serror = ehc->i.serror;
1047 unsigned int err_mask = 0, action = 0;
1049 if (serror & SERR_PERSISTENT) {
1050 err_mask |= AC_ERR_ATA_BUS;
1051 action |= ATA_EH_HARDRESET;
1054 (SERR_DATA_RECOVERED | SERR_COMM_RECOVERED | SERR_DATA)) {
1055 err_mask |= AC_ERR_ATA_BUS;
1056 action |= ATA_EH_SOFTRESET;
1058 if (serror & SERR_PROTOCOL) {
1059 err_mask |= AC_ERR_HSM;
1060 action |= ATA_EH_SOFTRESET;
1062 if (serror & SERR_INTERNAL) {
1063 err_mask |= AC_ERR_SYSTEM;
1064 action |= ATA_EH_HARDRESET;
1066 if (serror & (SERR_PHYRDY_CHG | SERR_DEV_XCHG))
1067 ata_ehi_hotplugged(&ehc->i);
1069 ehc->i.err_mask |= err_mask;
1070 ehc->i.action |= action;
1074 * ata_eh_analyze_ncq_error - analyze NCQ error
1075 * @ap: ATA port to analyze NCQ error for
1077 * Read log page 10h, determine the offending qc and acquire
1078 * error status TF. For NCQ device errors, all LLDDs have to do
1079 * is setting AC_ERR_DEV in ehi->err_mask. This function takes
1083 * Kernel thread context (may sleep).
1085 static void ata_eh_analyze_ncq_error(struct ata_port *ap)
1087 struct ata_eh_context *ehc = &ap->eh_context;
1088 struct ata_device *dev = ap->device;
1089 struct ata_queued_cmd *qc;
1090 struct ata_taskfile tf;
1093 /* if frozen, we can't do much */
1094 if (ap->pflags & ATA_PFLAG_FROZEN)
1097 /* is it NCQ device error? */
1098 if (!ap->sactive || !(ehc->i.err_mask & AC_ERR_DEV))
1101 /* has LLDD analyzed already? */
1102 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1103 qc = __ata_qc_from_tag(ap, tag);
1105 if (!(qc->flags & ATA_QCFLAG_FAILED))
1112 /* okay, this error is ours */
1113 rc = ata_eh_read_log_10h(dev, &tag, &tf);
1115 ata_port_printk(ap, KERN_ERR, "failed to read log page 10h "
1116 "(errno=%d)\n", rc);
1120 if (!(ap->sactive & (1 << tag))) {
1121 ata_port_printk(ap, KERN_ERR, "log page 10h reported "
1122 "inactive tag %d\n", tag);
1126 /* we've got the perpetrator, condemn it */
1127 qc = __ata_qc_from_tag(ap, tag);
1128 memcpy(&qc->result_tf, &tf, sizeof(tf));
1129 qc->err_mask |= AC_ERR_DEV;
1130 ehc->i.err_mask &= ~AC_ERR_DEV;
1134 * ata_eh_analyze_tf - analyze taskfile of a failed qc
1135 * @qc: qc to analyze
1136 * @tf: Taskfile registers to analyze
1138 * Analyze taskfile of @qc and further determine cause of
1139 * failure. This function also requests ATAPI sense data if
1143 * Kernel thread context (may sleep).
1146 * Determined recovery action
1148 static unsigned int ata_eh_analyze_tf(struct ata_queued_cmd *qc,
1149 const struct ata_taskfile *tf)
1151 unsigned int tmp, action = 0;
1152 u8 stat = tf->command, err = tf->feature;
1154 if ((stat & (ATA_BUSY | ATA_DRQ | ATA_DRDY)) != ATA_DRDY) {
1155 qc->err_mask |= AC_ERR_HSM;
1156 return ATA_EH_SOFTRESET;
1159 if (stat & (ATA_ERR | ATA_DF))
1160 qc->err_mask |= AC_ERR_DEV;
1164 switch (qc->dev->class) {
1167 qc->err_mask |= AC_ERR_ATA_BUS;
1169 qc->err_mask |= AC_ERR_MEDIA;
1171 qc->err_mask |= AC_ERR_INVALID;
1175 if (!(qc->ap->pflags & ATA_PFLAG_FROZEN)) {
1176 tmp = atapi_eh_request_sense(qc);
1178 /* ATA_QCFLAG_SENSE_VALID is used to
1179 * tell atapi_qc_complete() that sense
1180 * data is already valid.
1182 * TODO: interpret sense data and set
1183 * appropriate err_mask.
1185 qc->flags |= ATA_QCFLAG_SENSE_VALID;
1187 qc->err_mask |= tmp;
1191 if (qc->err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT | AC_ERR_ATA_BUS))
1192 action |= ATA_EH_SOFTRESET;
1197 static int ata_eh_categorize_error(int is_io, unsigned int err_mask)
1199 if (err_mask & AC_ERR_ATA_BUS)
1202 if (err_mask & AC_ERR_TIMEOUT)
1206 if (err_mask & AC_ERR_HSM)
1209 (AC_ERR_DEV|AC_ERR_MEDIA|AC_ERR_INVALID)) == AC_ERR_DEV)
1216 struct speed_down_verdict_arg {
1221 static int speed_down_verdict_cb(struct ata_ering_entry *ent, void *void_arg)
1223 struct speed_down_verdict_arg *arg = void_arg;
1224 int cat = ata_eh_categorize_error(ent->is_io, ent->err_mask);
1226 if (ent->timestamp < arg->since)
1229 arg->nr_errors[cat]++;
1234 * ata_eh_speed_down_verdict - Determine speed down verdict
1235 * @dev: Device of interest
1237 * This function examines error ring of @dev and determines
1238 * whether NCQ needs to be turned off, transfer speed should be
1239 * stepped down, or falling back to PIO is necessary.
1241 * Cat-1 is ATA_BUS error for any command.
1243 * Cat-2 is TIMEOUT for any command or HSM violation for known
1244 * supported commands.
1246 * Cat-3 is is unclassified DEV error for known supported
1249 * NCQ needs to be turned off if there have been more than 3
1250 * Cat-2 + Cat-3 errors during last 10 minutes.
1252 * Speed down is necessary if there have been more than 3 Cat-1 +
1253 * Cat-2 errors or 10 Cat-3 errors during last 10 minutes.
1255 * Falling back to PIO mode is necessary if there have been more
1256 * than 10 Cat-1 + Cat-2 + Cat-3 errors during last 5 minutes.
1259 * Inherited from caller.
1262 * OR of ATA_EH_SPDN_* flags.
1264 static unsigned int ata_eh_speed_down_verdict(struct ata_device *dev)
1266 const u64 j5mins = 5LLU * 60 * HZ, j10mins = 10LLU * 60 * HZ;
1267 u64 j64 = get_jiffies_64();
1268 struct speed_down_verdict_arg arg;
1269 unsigned int verdict = 0;
1271 /* scan past 10 mins of error history */
1272 memset(&arg, 0, sizeof(arg));
1273 arg.since = j64 - min(j64, j10mins);
1274 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1276 if (arg.nr_errors[2] + arg.nr_errors[3] > 3)
1277 verdict |= ATA_EH_SPDN_NCQ_OFF;
1278 if (arg.nr_errors[1] + arg.nr_errors[2] > 3 || arg.nr_errors[3] > 10)
1279 verdict |= ATA_EH_SPDN_SPEED_DOWN;
1281 /* scan past 3 mins of error history */
1282 memset(&arg, 0, sizeof(arg));
1283 arg.since = j64 - min(j64, j5mins);
1284 ata_ering_map(&dev->ering, speed_down_verdict_cb, &arg);
1286 if (arg.nr_errors[1] + arg.nr_errors[2] + arg.nr_errors[3] > 10)
1287 verdict |= ATA_EH_SPDN_FALLBACK_TO_PIO;
1293 * ata_eh_speed_down - record error and speed down if necessary
1294 * @dev: Failed device
1295 * @is_io: Did the device fail during normal IO?
1296 * @err_mask: err_mask of the error
1298 * Record error and examine error history to determine whether
1299 * adjusting transmission speed is necessary. It also sets
1300 * transmission limits appropriately if such adjustment is
1304 * Kernel thread context (may sleep).
1307 * Determined recovery action.
1309 static unsigned int ata_eh_speed_down(struct ata_device *dev, int is_io,
1310 unsigned int err_mask)
1312 unsigned int verdict;
1313 unsigned int action = 0;
1315 /* don't bother if Cat-0 error */
1316 if (ata_eh_categorize_error(is_io, err_mask) == 0)
1319 /* record error and determine whether speed down is necessary */
1320 ata_ering_record(&dev->ering, is_io, err_mask);
1321 verdict = ata_eh_speed_down_verdict(dev);
1324 if ((verdict & ATA_EH_SPDN_NCQ_OFF) &&
1325 (dev->flags & (ATA_DFLAG_PIO | ATA_DFLAG_NCQ |
1326 ATA_DFLAG_NCQ_OFF)) == ATA_DFLAG_NCQ) {
1327 dev->flags |= ATA_DFLAG_NCQ_OFF;
1328 ata_dev_printk(dev, KERN_WARNING,
1329 "NCQ disabled due to excessive errors\n");
1334 if (verdict & ATA_EH_SPDN_SPEED_DOWN) {
1335 /* speed down SATA link speed if possible */
1336 if (sata_down_spd_limit(dev->ap) == 0) {
1337 action |= ATA_EH_HARDRESET;
1341 /* lower transfer mode */
1342 if (dev->spdn_cnt < 2) {
1343 static const int dma_dnxfer_sel[] =
1344 { ATA_DNXFER_DMA, ATA_DNXFER_40C };
1345 static const int pio_dnxfer_sel[] =
1346 { ATA_DNXFER_PIO, ATA_DNXFER_FORCE_PIO0 };
1349 if (dev->xfer_shift != ATA_SHIFT_PIO)
1350 sel = dma_dnxfer_sel[dev->spdn_cnt];
1352 sel = pio_dnxfer_sel[dev->spdn_cnt];
1356 if (ata_down_xfermask_limit(dev, sel) == 0) {
1357 action |= ATA_EH_SOFTRESET;
1363 /* Fall back to PIO? Slowing down to PIO is meaningless for
1364 * SATA. Consider it only for PATA.
1366 if ((verdict & ATA_EH_SPDN_FALLBACK_TO_PIO) && (dev->spdn_cnt >= 2) &&
1367 (dev->ap->cbl != ATA_CBL_SATA) &&
1368 (dev->xfer_shift != ATA_SHIFT_PIO)) {
1369 if (ata_down_xfermask_limit(dev, ATA_DNXFER_FORCE_PIO) == 0) {
1371 action |= ATA_EH_SOFTRESET;
1378 /* device has been slowed down, blow error history */
1379 ata_ering_clear(&dev->ering);
1384 * ata_eh_autopsy - analyze error and determine recovery action
1385 * @ap: ATA port to perform autopsy on
1387 * Analyze why @ap failed and determine which recovery action is
1388 * needed. This function also sets more detailed AC_ERR_* values
1389 * and fills sense data for ATAPI CHECK SENSE.
1392 * Kernel thread context (may sleep).
1394 static void ata_eh_autopsy(struct ata_port *ap)
1396 struct ata_eh_context *ehc = &ap->eh_context;
1397 unsigned int all_err_mask = 0;
1404 if (ehc->i.flags & ATA_EHI_NO_AUTOPSY)
1407 /* obtain and analyze SError */
1408 rc = sata_scr_read(ap, SCR_ERROR, &serror);
1410 ehc->i.serror |= serror;
1411 ata_eh_analyze_serror(ap);
1412 } else if (rc != -EOPNOTSUPP)
1413 ehc->i.action |= ATA_EH_HARDRESET;
1415 /* analyze NCQ failure */
1416 ata_eh_analyze_ncq_error(ap);
1418 /* any real error trumps AC_ERR_OTHER */
1419 if (ehc->i.err_mask & ~AC_ERR_OTHER)
1420 ehc->i.err_mask &= ~AC_ERR_OTHER;
1422 all_err_mask |= ehc->i.err_mask;
1424 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1425 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1427 if (!(qc->flags & ATA_QCFLAG_FAILED))
1430 /* inherit upper level err_mask */
1431 qc->err_mask |= ehc->i.err_mask;
1434 ehc->i.action |= ata_eh_analyze_tf(qc, &qc->result_tf);
1436 /* DEV errors are probably spurious in case of ATA_BUS error */
1437 if (qc->err_mask & AC_ERR_ATA_BUS)
1438 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_MEDIA |
1441 /* any real error trumps unknown error */
1442 if (qc->err_mask & ~AC_ERR_OTHER)
1443 qc->err_mask &= ~AC_ERR_OTHER;
1445 /* SENSE_VALID trumps dev/unknown error and revalidation */
1446 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
1447 qc->err_mask &= ~(AC_ERR_DEV | AC_ERR_OTHER);
1448 ehc->i.action &= ~ATA_EH_REVALIDATE;
1451 /* accumulate error info */
1452 ehc->i.dev = qc->dev;
1453 all_err_mask |= qc->err_mask;
1454 if (qc->flags & ATA_QCFLAG_IO)
1458 /* enforce default EH actions */
1459 if (ap->pflags & ATA_PFLAG_FROZEN ||
1460 all_err_mask & (AC_ERR_HSM | AC_ERR_TIMEOUT))
1461 ehc->i.action |= ATA_EH_SOFTRESET;
1462 else if (all_err_mask)
1463 ehc->i.action |= ATA_EH_REVALIDATE;
1465 /* if we have offending qcs and the associated failed device */
1468 ehc->i.action |= ata_eh_speed_down(ehc->i.dev, is_io,
1471 /* perform per-dev EH action only on the offending device */
1472 ehc->i.dev_action[ehc->i.dev->devno] |=
1473 ehc->i.action & ATA_EH_PERDEV_MASK;
1474 ehc->i.action &= ~ATA_EH_PERDEV_MASK;
1481 * ata_eh_report - report error handling to user
1482 * @ap: ATA port EH is going on
1484 * Report EH to user.
1489 static void ata_eh_report(struct ata_port *ap)
1491 struct ata_eh_context *ehc = &ap->eh_context;
1492 const char *frozen, *desc;
1493 int tag, nr_failed = 0;
1496 if (ehc->i.desc[0] != '\0')
1499 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1500 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1502 if (!(qc->flags & ATA_QCFLAG_FAILED))
1504 if (qc->flags & ATA_QCFLAG_SENSE_VALID && !qc->err_mask)
1510 if (!nr_failed && !ehc->i.err_mask)
1514 if (ap->pflags & ATA_PFLAG_FROZEN)
1518 ata_dev_printk(ehc->i.dev, KERN_ERR, "exception Emask 0x%x "
1519 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1520 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1521 ehc->i.action, frozen);
1523 ata_dev_printk(ehc->i.dev, KERN_ERR, "(%s)\n", desc);
1525 ata_port_printk(ap, KERN_ERR, "exception Emask 0x%x "
1526 "SAct 0x%x SErr 0x%x action 0x%x%s\n",
1527 ehc->i.err_mask, ap->sactive, ehc->i.serror,
1528 ehc->i.action, frozen);
1530 ata_port_printk(ap, KERN_ERR, "(%s)\n", desc);
1533 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
1534 static const char *dma_str[] = {
1535 [DMA_BIDIRECTIONAL] = "bidi",
1536 [DMA_TO_DEVICE] = "out",
1537 [DMA_FROM_DEVICE] = "in",
1540 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
1541 struct ata_taskfile *cmd = &qc->tf, *res = &qc->result_tf;
1543 if (!(qc->flags & ATA_QCFLAG_FAILED) || !qc->err_mask)
1546 ata_dev_printk(qc->dev, KERN_ERR,
1547 "cmd %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1548 "tag %d cdb 0x%x data %u %s\n "
1549 "res %02x/%02x:%02x:%02x:%02x:%02x/%02x:%02x:%02x:%02x:%02x/%02x "
1550 "Emask 0x%x (%s)\n",
1551 cmd->command, cmd->feature, cmd->nsect,
1552 cmd->lbal, cmd->lbam, cmd->lbah,
1553 cmd->hob_feature, cmd->hob_nsect,
1554 cmd->hob_lbal, cmd->hob_lbam, cmd->hob_lbah,
1555 cmd->device, qc->tag, qc->cdb[0], qc->nbytes,
1556 dma_str[qc->dma_dir],
1557 res->command, res->feature, res->nsect,
1558 res->lbal, res->lbam, res->lbah,
1559 res->hob_feature, res->hob_nsect,
1560 res->hob_lbal, res->hob_lbam, res->hob_lbah,
1561 res->device, qc->err_mask, ata_err_string(qc->err_mask));
1565 static int ata_do_reset(struct ata_port *ap, ata_reset_fn_t reset,
1566 unsigned int *classes, unsigned long deadline)
1570 for (i = 0; i < ATA_MAX_DEVICES; i++)
1571 classes[i] = ATA_DEV_UNKNOWN;
1573 rc = reset(ap, classes, deadline);
1577 /* If any class isn't ATA_DEV_UNKNOWN, consider classification
1578 * is complete and convert all ATA_DEV_UNKNOWN to
1581 for (i = 0; i < ATA_MAX_DEVICES; i++)
1582 if (classes[i] != ATA_DEV_UNKNOWN)
1585 if (i < ATA_MAX_DEVICES)
1586 for (i = 0; i < ATA_MAX_DEVICES; i++)
1587 if (classes[i] == ATA_DEV_UNKNOWN)
1588 classes[i] = ATA_DEV_NONE;
1593 static int ata_eh_followup_srst_needed(int rc, int classify,
1594 const unsigned int *classes)
1600 if (classify && classes[0] == ATA_DEV_UNKNOWN)
1605 static int ata_eh_reset(struct ata_port *ap, int classify,
1606 ata_prereset_fn_t prereset, ata_reset_fn_t softreset,
1607 ata_reset_fn_t hardreset, ata_postreset_fn_t postreset)
1609 struct ata_eh_context *ehc = &ap->eh_context;
1610 unsigned int *classes = ehc->classes;
1611 int verbose = !(ehc->i.flags & ATA_EHI_QUIET);
1613 unsigned long deadline;
1614 unsigned int action;
1615 ata_reset_fn_t reset;
1616 int i, did_followup_srst, rc;
1618 /* about to reset */
1619 ata_eh_about_to_do(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1621 /* Determine which reset to use and record in ehc->i.action.
1622 * prereset() may examine and modify it.
1624 action = ehc->i.action;
1625 ehc->i.action &= ~ATA_EH_RESET_MASK;
1626 if (softreset && (!hardreset || (!sata_set_spd_needed(ap) &&
1627 !(action & ATA_EH_HARDRESET))))
1628 ehc->i.action |= ATA_EH_SOFTRESET;
1630 ehc->i.action |= ATA_EH_HARDRESET;
1633 rc = prereset(ap, jiffies + ATA_EH_PRERESET_TIMEOUT);
1635 if (rc == -ENOENT) {
1636 ata_port_printk(ap, KERN_DEBUG,
1637 "port disabled. ignoring.\n");
1638 ap->eh_context.i.action &= ~ATA_EH_RESET_MASK;
1640 for (i = 0; i < ATA_MAX_DEVICES; i++)
1641 classes[i] = ATA_DEV_NONE;
1645 ata_port_printk(ap, KERN_ERR,
1646 "prereset failed (errno=%d)\n", rc);
1651 /* prereset() might have modified ehc->i.action */
1652 if (ehc->i.action & ATA_EH_HARDRESET)
1654 else if (ehc->i.action & ATA_EH_SOFTRESET)
1657 /* prereset told us not to reset, bang classes and return */
1658 for (i = 0; i < ATA_MAX_DEVICES; i++)
1659 classes[i] = ATA_DEV_NONE;
1663 /* did prereset() screw up? if so, fix up to avoid oopsing */
1665 ata_port_printk(ap, KERN_ERR, "BUG: prereset() requested "
1666 "invalid reset type\n");
1674 deadline = jiffies + ata_eh_reset_timeouts[try++];
1676 /* shut up during boot probing */
1678 ata_port_printk(ap, KERN_INFO, "%s resetting port\n",
1679 reset == softreset ? "soft" : "hard");
1681 /* mark that this EH session started with reset */
1682 if (reset == hardreset)
1683 ehc->i.flags |= ATA_EHI_DID_HARDRESET;
1685 ehc->i.flags |= ATA_EHI_DID_SOFTRESET;
1687 rc = ata_do_reset(ap, reset, classes, deadline);
1689 did_followup_srst = 0;
1690 if (reset == hardreset &&
1691 ata_eh_followup_srst_needed(rc, classify, classes)) {
1692 /* okay, let's do follow-up softreset */
1693 did_followup_srst = 1;
1697 ata_port_printk(ap, KERN_ERR,
1698 "follow-up softreset required "
1699 "but no softreset avaliable\n");
1703 ata_eh_about_to_do(ap, NULL, ATA_EH_RESET_MASK);
1704 rc = ata_do_reset(ap, reset, classes, deadline);
1706 if (rc == 0 && classify &&
1707 classes[0] == ATA_DEV_UNKNOWN) {
1708 ata_port_printk(ap, KERN_ERR,
1709 "classification failed\n");
1714 if (rc && try < ARRAY_SIZE(ata_eh_reset_timeouts)) {
1715 unsigned long now = jiffies;
1717 if (time_before(now, deadline)) {
1718 unsigned long delta = deadline - jiffies;
1720 ata_port_printk(ap, KERN_WARNING, "reset failed "
1721 "(errno=%d), retrying in %u secs\n",
1722 rc, (jiffies_to_msecs(delta) + 999) / 1000);
1724 schedule_timeout_uninterruptible(delta);
1727 if (reset == hardreset &&
1728 try == ARRAY_SIZE(ata_eh_reset_timeouts) - 1)
1729 sata_down_spd_limit(ap);
1736 /* After the reset, the device state is PIO 0 and the
1737 * controller state is undefined. Record the mode.
1739 for (i = 0; i < ATA_MAX_DEVICES; i++)
1740 ap->device[i].pio_mode = XFER_PIO_0;
1743 postreset(ap, classes);
1745 /* reset successful, schedule revalidation */
1746 ata_eh_done(ap, NULL, ehc->i.action & ATA_EH_RESET_MASK);
1747 ehc->i.action |= ATA_EH_REVALIDATE;
1753 static int ata_eh_revalidate_and_attach(struct ata_port *ap,
1754 struct ata_device **r_failed_dev)
1756 struct ata_eh_context *ehc = &ap->eh_context;
1757 struct ata_device *dev;
1758 unsigned int new_mask = 0;
1759 unsigned long flags;
1764 /* For PATA drive side cable detection to work, IDENTIFY must
1765 * be done backwards such that PDIAG- is released by the slave
1766 * device before the master device is identified.
1768 for (i = ATA_MAX_DEVICES - 1; i >= 0; i--) {
1769 unsigned int action, readid_flags = 0;
1771 dev = &ap->device[i];
1772 action = ata_eh_dev_action(dev);
1774 if (ehc->i.flags & ATA_EHI_DID_RESET)
1775 readid_flags |= ATA_READID_POSTRESET;
1777 if ((action & ATA_EH_REVALIDATE) && ata_dev_enabled(dev)) {
1778 if (ata_port_offline(ap)) {
1783 ata_eh_about_to_do(ap, dev, ATA_EH_REVALIDATE);
1784 rc = ata_dev_revalidate(dev, readid_flags);
1788 ata_eh_done(ap, dev, ATA_EH_REVALIDATE);
1790 /* Configuration may have changed, reconfigure
1793 ehc->i.flags |= ATA_EHI_SETMODE;
1795 /* schedule the scsi_rescan_device() here */
1796 queue_work(ata_aux_wq, &(ap->scsi_rescan_task));
1797 } else if (dev->class == ATA_DEV_UNKNOWN &&
1798 ehc->tries[dev->devno] &&
1799 ata_class_enabled(ehc->classes[dev->devno])) {
1800 dev->class = ehc->classes[dev->devno];
1802 rc = ata_dev_read_id(dev, &dev->class, readid_flags,
1809 /* IDENTIFY was issued to non-existent
1810 * device. No need to reset. Just
1811 * thaw and kill the device.
1813 ata_eh_thaw_port(ap);
1814 dev->class = ATA_DEV_UNKNOWN;
1817 dev->class = ATA_DEV_UNKNOWN;
1823 /* PDIAG- should have been released, ask cable type if post-reset */
1824 if ((ehc->i.flags & ATA_EHI_DID_RESET) && ap->ops->cable_detect)
1825 ap->cbl = ap->ops->cable_detect(ap);
1827 /* Configure new devices forward such that user doesn't see
1828 * device detection messages backwards.
1830 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1831 dev = &ap->device[i];
1833 if (!(new_mask & (1 << i)))
1836 ehc->i.flags |= ATA_EHI_PRINTINFO;
1837 rc = ata_dev_configure(dev);
1838 ehc->i.flags &= ~ATA_EHI_PRINTINFO;
1842 spin_lock_irqsave(ap->lock, flags);
1843 ap->pflags |= ATA_PFLAG_SCSI_HOTPLUG;
1844 spin_unlock_irqrestore(ap->lock, flags);
1846 /* new device discovered, configure xfermode */
1847 ehc->i.flags |= ATA_EHI_SETMODE;
1853 *r_failed_dev = dev;
1854 DPRINTK("EXIT rc=%d\n", rc);
1858 static int ata_port_nr_enabled(struct ata_port *ap)
1862 for (i = 0; i < ATA_MAX_DEVICES; i++)
1863 if (ata_dev_enabled(&ap->device[i]))
1868 static int ata_port_nr_vacant(struct ata_port *ap)
1872 for (i = 0; i < ATA_MAX_DEVICES; i++)
1873 if (ap->device[i].class == ATA_DEV_UNKNOWN)
1878 static int ata_eh_skip_recovery(struct ata_port *ap)
1880 struct ata_eh_context *ehc = &ap->eh_context;
1883 /* thaw frozen port, resume link and recover failed devices */
1884 if ((ap->pflags & ATA_PFLAG_FROZEN) ||
1885 (ehc->i.flags & ATA_EHI_RESUME_LINK) || ata_port_nr_enabled(ap))
1888 /* skip if class codes for all vacant slots are ATA_DEV_NONE */
1889 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1890 struct ata_device *dev = &ap->device[i];
1892 if (dev->class == ATA_DEV_UNKNOWN &&
1893 ehc->classes[dev->devno] != ATA_DEV_NONE)
1901 * ata_eh_recover - recover host port after error
1902 * @ap: host port to recover
1903 * @prereset: prereset method (can be NULL)
1904 * @softreset: softreset method (can be NULL)
1905 * @hardreset: hardreset method (can be NULL)
1906 * @postreset: postreset method (can be NULL)
1908 * This is the alpha and omega, eum and yang, heart and soul of
1909 * libata exception handling. On entry, actions required to
1910 * recover the port and hotplug requests are recorded in
1911 * eh_context. This function executes all the operations with
1912 * appropriate retrials and fallbacks to resurrect failed
1913 * devices, detach goners and greet newcomers.
1916 * Kernel thread context (may sleep).
1919 * 0 on success, -errno on failure.
1921 static int ata_eh_recover(struct ata_port *ap, ata_prereset_fn_t prereset,
1922 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
1923 ata_postreset_fn_t postreset)
1925 struct ata_eh_context *ehc = &ap->eh_context;
1926 struct ata_device *dev;
1931 /* prep for recovery */
1932 for (i = 0; i < ATA_MAX_DEVICES; i++) {
1933 dev = &ap->device[i];
1935 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
1937 /* collect port action mask recorded in dev actions */
1938 ehc->i.action |= ehc->i.dev_action[i] & ~ATA_EH_PERDEV_MASK;
1939 ehc->i.dev_action[i] &= ATA_EH_PERDEV_MASK;
1941 /* process hotplug request */
1942 if (dev->flags & ATA_DFLAG_DETACH)
1943 ata_eh_detach_dev(dev);
1945 if (!ata_dev_enabled(dev) &&
1946 ((ehc->i.probe_mask & (1 << dev->devno)) &&
1947 !(ehc->did_probe_mask & (1 << dev->devno)))) {
1948 ata_eh_detach_dev(dev);
1950 ehc->did_probe_mask |= (1 << dev->devno);
1951 ehc->i.action |= ATA_EH_SOFTRESET;
1958 /* if UNLOADING, finish immediately */
1959 if (ap->pflags & ATA_PFLAG_UNLOADING)
1962 /* skip EH if possible. */
1963 if (ata_eh_skip_recovery(ap))
1966 for (i = 0; i < ATA_MAX_DEVICES; i++)
1967 ehc->classes[i] = ATA_DEV_UNKNOWN;
1970 if (ehc->i.action & ATA_EH_RESET_MASK) {
1971 ata_eh_freeze_port(ap);
1973 rc = ata_eh_reset(ap, ata_port_nr_vacant(ap), prereset,
1974 softreset, hardreset, postreset);
1976 ata_port_printk(ap, KERN_ERR,
1977 "reset failed, giving up\n");
1981 ata_eh_thaw_port(ap);
1984 /* revalidate existing devices and attach new ones */
1985 rc = ata_eh_revalidate_and_attach(ap, &dev);
1989 /* configure transfer mode if necessary */
1990 if (ehc->i.flags & ATA_EHI_SETMODE) {
1991 rc = ata_set_mode(ap, &dev);
1994 ehc->i.flags &= ~ATA_EHI_SETMODE;
2000 ehc->tries[dev->devno]--;
2004 /* eeek, something went very wrong, give up */
2005 ehc->tries[dev->devno] = 0;
2009 /* device missing or wrong IDENTIFY data, schedule probing */
2010 ehc->i.probe_mask |= (1 << dev->devno);
2011 /* give it just one more chance */
2012 ehc->tries[dev->devno] = min(ehc->tries[dev->devno], 1);
2014 if (ehc->tries[dev->devno] == 1) {
2015 /* This is the last chance, better to slow
2016 * down than lose it.
2018 sata_down_spd_limit(ap);
2019 ata_down_xfermask_limit(dev, ATA_DNXFER_PIO);
2023 if (ata_dev_enabled(dev) && !ehc->tries[dev->devno]) {
2024 /* disable device if it has used up all its chances */
2025 ata_dev_disable(dev);
2027 /* detach if offline */
2028 if (ata_port_offline(ap))
2029 ata_eh_detach_dev(dev);
2031 /* probe if requested */
2032 if ((ehc->i.probe_mask & (1 << dev->devno)) &&
2033 !(ehc->did_probe_mask & (1 << dev->devno))) {
2034 ata_eh_detach_dev(dev);
2037 ehc->tries[dev->devno] = ATA_EH_DEV_TRIES;
2038 ehc->did_probe_mask |= (1 << dev->devno);
2039 ehc->i.action |= ATA_EH_SOFTRESET;
2042 /* soft didn't work? be haaaaard */
2043 if (ehc->i.flags & ATA_EHI_DID_RESET)
2044 ehc->i.action |= ATA_EH_HARDRESET;
2046 ehc->i.action |= ATA_EH_SOFTRESET;
2049 if (ata_port_nr_enabled(ap)) {
2050 ata_port_printk(ap, KERN_WARNING, "failed to recover some "
2051 "devices, retrying in 5 secs\n");
2054 /* no device left, repeat fast */
2062 for (i = 0; i < ATA_MAX_DEVICES; i++)
2063 ata_dev_disable(&ap->device[i]);
2066 DPRINTK("EXIT, rc=%d\n", rc);
2071 * ata_eh_finish - finish up EH
2072 * @ap: host port to finish EH for
2074 * Recovery is complete. Clean up EH states and retry or finish
2080 static void ata_eh_finish(struct ata_port *ap)
2084 /* retry or finish qcs */
2085 for (tag = 0; tag < ATA_MAX_QUEUE; tag++) {
2086 struct ata_queued_cmd *qc = __ata_qc_from_tag(ap, tag);
2088 if (!(qc->flags & ATA_QCFLAG_FAILED))
2092 /* FIXME: Once EH migration is complete,
2093 * generate sense data in this function,
2094 * considering both err_mask and tf.
2096 if (qc->err_mask & AC_ERR_INVALID)
2097 ata_eh_qc_complete(qc);
2099 ata_eh_qc_retry(qc);
2101 if (qc->flags & ATA_QCFLAG_SENSE_VALID) {
2102 ata_eh_qc_complete(qc);
2104 /* feed zero TF to sense generation */
2105 memset(&qc->result_tf, 0, sizeof(qc->result_tf));
2106 ata_eh_qc_retry(qc);
2113 * ata_do_eh - do standard error handling
2114 * @ap: host port to handle error for
2115 * @prereset: prereset method (can be NULL)
2116 * @softreset: softreset method (can be NULL)
2117 * @hardreset: hardreset method (can be NULL)
2118 * @postreset: postreset method (can be NULL)
2120 * Perform standard error handling sequence.
2123 * Kernel thread context (may sleep).
2125 void ata_do_eh(struct ata_port *ap, ata_prereset_fn_t prereset,
2126 ata_reset_fn_t softreset, ata_reset_fn_t hardreset,
2127 ata_postreset_fn_t postreset)
2131 ata_eh_recover(ap, prereset, softreset, hardreset, postreset);
2137 * ata_eh_handle_port_suspend - perform port suspend operation
2138 * @ap: port to suspend
2143 * Kernel thread context (may sleep).
2145 static void ata_eh_handle_port_suspend(struct ata_port *ap)
2147 unsigned long flags;
2150 /* are we suspending? */
2151 spin_lock_irqsave(ap->lock, flags);
2152 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2153 ap->pm_mesg.event == PM_EVENT_ON) {
2154 spin_unlock_irqrestore(ap->lock, flags);
2157 spin_unlock_irqrestore(ap->lock, flags);
2159 WARN_ON(ap->pflags & ATA_PFLAG_SUSPENDED);
2162 ata_eh_freeze_port(ap);
2164 if (ap->ops->port_suspend)
2165 rc = ap->ops->port_suspend(ap, ap->pm_mesg);
2168 spin_lock_irqsave(ap->lock, flags);
2170 ap->pflags &= ~ATA_PFLAG_PM_PENDING;
2172 ap->pflags |= ATA_PFLAG_SUSPENDED;
2174 ata_port_schedule_eh(ap);
2176 if (ap->pm_result) {
2177 *ap->pm_result = rc;
2178 ap->pm_result = NULL;
2181 spin_unlock_irqrestore(ap->lock, flags);
2187 * ata_eh_handle_port_resume - perform port resume operation
2188 * @ap: port to resume
2193 * Kernel thread context (may sleep).
2195 static void ata_eh_handle_port_resume(struct ata_port *ap)
2197 unsigned long flags;
2200 /* are we resuming? */
2201 spin_lock_irqsave(ap->lock, flags);
2202 if (!(ap->pflags & ATA_PFLAG_PM_PENDING) ||
2203 ap->pm_mesg.event != PM_EVENT_ON) {
2204 spin_unlock_irqrestore(ap->lock, flags);
2207 spin_unlock_irqrestore(ap->lock, flags);
2209 WARN_ON(!(ap->pflags & ATA_PFLAG_SUSPENDED));
2211 if (ap->ops->port_resume)
2212 rc = ap->ops->port_resume(ap);
2215 spin_lock_irqsave(ap->lock, flags);
2216 ap->pflags &= ~(ATA_PFLAG_PM_PENDING | ATA_PFLAG_SUSPENDED);
2217 if (ap->pm_result) {
2218 *ap->pm_result = rc;
2219 ap->pm_result = NULL;
2221 spin_unlock_irqrestore(ap->lock, flags);
2223 #endif /* CONFIG_PM */